A unified account-based AI conversation governance method and device

By setting up member directories and metering ledgers for family or organization accounts, when a terminal initiates a session, it determines the local or cloud governance point based on the reachability of the local governance anchor point. This solves the problems of consistency in governance point selection and lossless upgrades and downgrades between different forms under the same account, and achieves seamless switching between privacy protection and metering memory.

CN123122826APending Publication Date: 2026-11-17廖长林
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611454552.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-18
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Under the same family or organization account, there is a lack of a solution that enables sessions to deterministically select the governance point based on account type and whether the terminal is on the local network, ensuring consistency of member policies, access credentials, quotas, metering ledgers and memories between two points, preventing privacy protection levels from being reduced during switching, and enabling seamless upgrades and downgrades of accounts between pure cloud mode and anchor-bound collaborative mode.

Method used

Set up a unique member directory, quota, and metering ledger for family or organization accounts. When a terminal initiates a session, it determines presence based on the reachability of the local governance anchor point and deterministically selects either the local or cloud governance point. Both points share the same set of member-based policies, and metering and memory are merged into the same ledger and memory domain. Administrators can configure governance policies for highly restricted members.

Benefits of technology

It enables members to use the same account and the same policy at home and outside, without downgrading privacy protection, without duplication of measurement memory, and without loss of upgrade or downgrade between account types, thus balancing high-level privacy protection with flexible governance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure FT_1
    Figure FT_1
  • Figure FT_2
    Figure FT_2
  • Figure FT_3
    Figure FT_3
Patent Text Reader

Abstract

This invention discloses a unified AI session management method and apparatus for accounts. When an account is not bound to an anchor point, it operates purely in the cloud. After binding to a broadband entry anchor point with a local trust root sealing policy and a recovery key, local access is prioritized by default. Members outside the network are automatically reconnected to the local network. Ordinary members can choose the cloud with administrator authorization, while highly restricted members do not automatically choose the ordinary cloud. Minors are pre-selected by their guardians, and confidential personnel do not revert to the cloud. The cloud path terminates at the aggregation entry point and is naturally visible at the application layer, based on informed consent, self-destruction after use, and no training. Both points apply the same set of connection-level and content-level policies. Privacy levels are verified before switching, and silent downgrading is avoided. Metering and memory are idempotently merged into the same ledger and memory domain using member, device, and session identifiers, outputting local, cloud, and unmanaged three-state tags. Accounts can seamlessly upgrade or downgrade between pure cloud and bound modes. Terminals without installed programs are managed by the anchor point according to network layer evidence connection level.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of edge computing, cloud-edge collaboration, and artificial intelligence governance technology. Specifically, it relates to a method and apparatus that allows cloud-based artificial intelligence governance points and local governance anchors deployed at broadband gateways to coexist under the same family or organization account, deterministically selects the governance point based on account type and whether the member terminal is on the local network, and ensures that member policies, credentials, quotas, metering ledgers, memories, and privacy protection levels remain consistent between the two governance points, and can be upgraded or downgraded seamlessly between pure cloud mode and local anchor mode. Background Technology

[0002] Existing cloud-based multi-model aggregation services provide developers and organizations with a unified interface, model routing, team workspaces, and quota allocation and usage statistics based on access credentials. Essentially, it is an AI call entry point located in the cloud, which can only manage requests that actively use its access credentials, its application, or its interface. It cannot reach terminals in home or organizational broadband access points that have not installed its program or configured its access credentials; and the session is terminated in plaintext at the cloud aggregation service. Existing home network security devices, combined with mobile applications, use security boxes connected in series or attached to the home network to perform domain name filtering, network control, and intrusion detection. The cloud is an auxiliary management surface of the box and is not a second governance point for artificial intelligence conversations. It does not undertake the desensitization of uplink sensitive information, restoration of downlink content according to member authorization level, or cross-large model memory aggregation. Existing enterprise security access services and zero-trust network access solutions place governance authority in the vendor's cloud, relying on persistent clients, enterprise identity systems, and dedicated operations and maintenance (O&M). This is unsuitable for families and small organizations without dedicated O&M support. Another type of existing cloud-based domain name and network filtering service maintains a single configuration for the same account. Its governance point is uniquely located in the cloud and operates only at the domain name resolution or connection layer. Home routers only handle domain name lookup and forwarding without holding restoration keys or acting as peer-to-peer content governance authority. One existing type of enterprise-level AI gateway supports various deployment models, including Software as a Service (SaaS), Virtual Private Cloud (VPN), self-hosting, and even air-gap deployment. It can keep sensitive paths within the customer's self-managed environment while entrusting other paths to managed cloud environments. It also provides zero data retention, data residency, virtual access credentials, and budget control. This type of solution is aimed at organizations with IT departments. Their local end is a cluster or VPN instance deployed and maintained by the customer themselves. It relies on the enterprise identity system and configures policies on both the local and cloud sides. It does not use a hardware root of trust to hold the restoration key, is plug-and-play, or cover broadband access anchors for families or small organizations without installed client terminals. It also does not provide consumers with an upgrade path that allows them to start with a pure cloud application with zero hardware and then seamlessly inherit the same account member directory, quota, ledger, and memory after scanning a code to bind the local anchor. One type of existing pure software AI service for families creates a profile for each member using a family account, masking personal identity information before message delivery models in its own application or cloud. The governance and masking of this type of solution both occur in its own application or cloud. Each terminal that needs protection has its own application installed and logged in. There is no governance anchor deployed at the broadband entry point and holding the restoration key with the local hardware root of trust. It also cannot cover smart TVs, game consoles, smart speakers and guest terminals that have not installed its application. Furthermore, there is no routing between the cloud governance point and the local anchor point based on the presence status when the terminal is at home or away, and there is no lossless upgrade and unbinding rollback after the pure software account is bound to the local anchor point. One type of existing home hub device undertakes local translation of smart device linkage within the local network, or distributes artificial intelligence computing power tasks between inference devices within the home and the cloud. Its processing objects are device control commands or computing power tasks. It does not perform content governance for artificial intelligence target sessions between the cloud governance point and broadband access anchor point of the same family or organization account, does not maintain the metering ledger and member memory domain that are unified with the governance point, and does not provide lossless upgrades or downgrades from zero-hardware pure cloud applications to the same account bound to the local hub. Therefore, when a family or small organization account uses both cloud-based AI aggregation and local broadband access governance as execution points, there is a lack of a solution that allows the session to deterministically select a governance point based on account type and whether the terminal is on a local network. This solution should ensure that both governance points apply consistent member-based policies to the same session, that privacy protection levels are not reduced during switching, that metering and memory are merged into the same ledger and memory domain to avoid duplicate billing, and that accounts can seamlessly upgrade between a pure cloud mode without a local anchor and a collaborative mode after anchor binding, and smoothly downgrade after anchor removal. Furthermore, after an account is bound to a local anchor, while members default to connecting locally when outside, there is a lack of a flexible governance point selection mechanism that can be configured by the administrator. This invention addresses the above shortcomings. Summary of the Invention

[0003] (1) Technical problems to be solved The technical problem this invention aims to solve is: when a family or organization account has both a cloud-based tenant governance point and a broadband entry point local governance anchor point, how can the AI ​​target session of a member deterministically select the governance point based on the account type and presence status, ensuring consistency of member policies, access credentials, quotas, metering ledgers, and memories between the two points, preventing the privacy protection level from being reduced due to the switching of governance points, and enabling seamless upgrades and downgrades of accounts between pure cloud mode and anchor-bound collaborative mode; furthermore, how can, after an account is bound to a local anchor point, members default to local connection when outside, while allowing ordinary members to choose cloud governance after pre-authorization by the administrator, while high-restriction members are always forced to connect to local; (2) Technical solution This invention sets up a unique member directory, quota, and metering ledger for a family or organization account. This account can exist in a pure cloud environment without being bound to a local governance anchor. Member terminals terminate and manage AI target sessions at the cloud tenant governance point via an access bearer program. After the account is bound to a local governance anchor deployed at the broadband gateway and sealed with a local root of trust, specifying the member policy instance and restoration key, the cloud only synchronizes the policy version (without the original sensitive values) and member pseudonym identifier to the anchor. When a member terminal initiates an AI target session, its presence is determined based on the reachability of the local governance anchor. Within the local network, the local governance anchor is used deterministically. When outside the coverage area of ​​the local governance anchor, the pure cloud environment defaults to using the cloud tenant governance point, while the anchor-bound environment defaults to using the local governance anchor, or selects cloud governance according to the pre-configured policy by the administrator. Both governance points apply the same membership-based policy, which includes two levels: connection level and content level. The governance point applies the appropriate level based on whether it can obtain application-layer plaintext for the session. When accessing the local system, the original value and restoration key remain within the local root of trust. When accessing the cloud system, the content to be governed terminates at the system's aggregation entry point, and the system connects to the terminal and model service via two independent connections. Application-layer content is naturally visible at the aggregation entry point. The metrics and memories generated by the two points are merged into the same ledger and memory domain using an idempotent key composed of member, device, and session identifiers, and deduplicated. For each session, a three-state flag is output: local governance, cloud governance, and ungoverned. After an account unbinds the anchor point, it reverts to pure cloud. When a pure cloud account binds an anchor point for the first time, members, policies, ledgers, and encrypted memories are inherited without loss. The pre-signature of credentials, public network blind relay and home idempotent takeover of the channel through which the member terminal reconnects to the local governance anchor point are carried out by the session roaming governance technology solution that is compatible with this application. This invention only specifies that the reconnection is one of the routing results that are parallel to cloud governance and local governance, and its selection relationship with account type, presence status and administrator authorization. Furthermore, after an account is bound to a local governance anchor point, the default routing policy switches to local priority, and when a member's terminal is outside the local area, it will default to routing back to the local governance anchor point. After pre-authorization by the administrator, ordinary users can choose to route AI target sessions to the cloud tenant governance point. Ordinary users who choose the cloud need the administrator to enable the allow switch in the management terminal or authorize it sequentially. High-restriction members are not allowed to choose the cloud on their own and are always forced to reconnect to the local governance anchor point. Administrators can configure differentiated allow policies for specified members, specified devices, or specified time periods in the management terminal. Furthermore, the highly restricted members are divided into two categories: the first category is minors, whose fallback strategy is preset by the guardian on the management end, which is selected from allowing access to the enhanced-filter cloud or controlled disconnection when the local area is unreachable; the second category is personnel bound by confidentiality obligations, whose fallback strategy is not to fall back to the ordinary cloud, but to use terminal desensitization or a trusted execution environment where the client controls the key. If neither of the two is feasible, the strictest snapshot is taken or the flow is controlled. (3) Beneficial effects First, because deterministic routing is used across local, cloud, and backhaul paths based on account type and presence status, members can use the same account and the same member-based policy whether at home or away, and it can cover all terminals in the house without requiring client installation when at home. Second, because the privacy protection level is determined by the governance point and the level is verified before switching and explicitly prompted during switching, the privacy protection will not be reduced without the user's awareness when switching governance points. Third, because the two-point metering and memory are merged into the same ledger and memory domain with idempotent keys and deduplicated, the same session will not be billed repeatedly or have its memory repeatedly stored when switching between the two points. Fourth, because the three-state markers distinguish between local governance and cloud governance. The system provides a status update for periods when the system is not visible, and provides a separate alert for periods of unknown status. Fifth, since accounts can seamlessly upgrade or downgrade between pure cloud and anchor-bound modes, users can start with cloud applications at zero hardware cost and add local anchors when whole-house coverage and local privacy are required. Sixth, since local access is prioritized by default after an account is bound to an anchor, ordinary members can only choose cloud governance after prior authorization by the administrator. High-restriction members are always forced to reconnect to local. For minors, guardians can preset enhanced cloud filtering or disconnection, and those bound by confidentiality obligations will not revert to the ordinary cloud, thus balancing high-level privacy protection with flexible governance needs. Attached Figure Description

[0004] Figure 1 A schematic diagram of the overall architecture for a unified dual governance point system for accounts; Figure 2 A flowchart for determining form and presence, and for deterministic route selection; Figure 3 A schematic diagram illustrating the idempotent fusion of three-state governance markers and ledger memory; Figure 4 This is a schematic diagram illustrating the lossless upgrade / downgrade process between the pure cloud form and the bound anchor point form. Figure 5 This is a schematic diagram illustrating how privacy protection levels are mapped to governance points. Figure 6 Flowchart for authorizing administrators to select between cloud and governance points. Detailed Implementation

[0005] (1) System composition and the same account system like Figure 1As shown, the account collaborative governance system consists of three types of components: a cloud tenant governance point located in the data center, an access bearing program on the member terminal, and a local governance anchor selectively bound to the same family or organization account. The local governance anchor is an access device deployed at the broadband entrance of a family or organization, and its hardware structure, selective traffic diversion, failure bypass, and forwarding governance isolation are supported by the in-transit governance integrated access device technical solution matching the present application; the local governance anchor derives mutually isolated sealed keys and sealing policy instances for the family or organization, each member and device group using the local trust root. The local trust root can be implemented by at least one of a hardware security module, one-time programmable memory or fuse, a trusted execution environment of a processor, an independent security element, and a protected software key store; The cloud tenant governance point maintains a unique member directory, quota and metering ledger for each family or organization account. Capabilities such as multi-model unified access, model routing based on cost and availability, failover, team workspace and quota allocation by access credentials belong to the existing cloud-side multi-model aggregation technology and are not the invention point of the present invention; the improvement of the present invention lies in how the cloud tenant governance point and the local governance anchor, as two governance points of the same account, cooperate. The access bearing program can be a web application, a native application, an operating system agent or a lightweight client. The cloud tenant governance point and the access bearing program are provided by the same service provider and configured in coordination, the local governance anchor is provided by the service provider or its authorized party, and the three are configured to cooperatively execute the method of the present invention; (2) Presence determination and morphology-related deterministic routing As shown in Figure 2 , before the member terminal initiates the artificial intelligence target session, the access bearing program or the local governance anchor performs presence determination. The presence evidence includes but is not limited to: the access bearing program receives the reachability identifier periodically advertised by the local governance anchor in the local network; the terminal and the anchor are in the same network address prefix or the same link layer broadcast domain; the online status reported by the anchor to the cloud through the egress long connection. At least two of the above evidences can be required to be consistent to determine that the terminal is present, and when the evidence is insufficient or contradictory, it is processed as not being within the coverage of the local governance anchor; The routing logic of the present invention is determined according to the account morphology and the presence status, and is divided into the following scenarios: First, in the pure cloud morphology where the account is not bound to the local governance anchor, the artificial intelligence target session of the member terminal is terminated and governed at the cloud tenant governance point through the access bearing program, and there is no local priority issue; Second, after the account is bound to the local governance anchor, when the terminal is in the local network, it is routed to the local governance point for governance; when the terminal is outside the local network, it is by default routed back to the local governance anchor for governance through the backhaul channel established by the session roaming governance solution matching the present application; Third, after an account is bound to a local governance anchor point, ordinary users can choose to route AI target sessions to the cloud tenant governance point for governance with prior authorization from the administrator; ordinary users who choose the cloud need the administrator to enable the allow switch or authorize it step by step in the management terminal; high-restriction members are not allowed to choose the cloud on their own and are always forced to reconnect to the local governance anchor point. For users who pre-define high-restriction members, sessions that require access to the memory domain of the member that exists only locally, or users who explicitly select strong consistency governance, the AI ​​target session will only be reconnected to the local governance anchor point via the reconnection channel established by the session roaming governance scheme that accompanies this application. If the local governance anchor point is unreachable, it will be processed according to the strict policy snapshot pre-signed before the anchor point went offline, or it will be routed to the cloud tenant governance point when the administrator pre-authorizes it. High-restriction members will be processed according to the two fallback methods in Section (8). If neither of these conditions is met, the session will be marked as unmanaged. After the account is bound to the local governance anchor point, the presence determination and connection-level management do not require each terminal to have the access bearer program installed: For terminals such as smart TVs, smart speakers, and game consoles that cannot or are inconvenient to install programs, their AI target sessions in the local network pass through the local governance anchor point at the network layer. The anchor point directly completes the presence determination and connection-level traffic diversion and management based on the fact that the terminal is in the local link layer broadcast domain or the same network address prefix and the session is transiting. The connection-level management referred to in this paper refers to the identification, access control, concurrency and frequency management and metering of AI target sessions based solely on connection information such as service identifiers, network addresses, protocol characteristics, 5-tuples and timing size, without decrypting application-layer plaintext; the content-level management referred to in this paper refers to the uplink sensitive value desensitization, downlink restoration by member authorization level, cross-model memory aggregation and unified credential forwarding when application-layer plaintext can be obtained. (3) Plaintext visibility and compliance boundaries of cloud paths This section clarifies the technical essence and compliance boundaries of cloud-based paths: First, the technical prerequisite for cloud-based content-level governance is that when governed by the cloud-based tenant governance point, the AI ​​target session terminates at the aggregation entry point provided by this system. This system then connects to the terminal and model services via two independent connections, with application-layer content naturally visible at the aggregation entry point. This visibility is a prerequisite for implementing content-level anonymization, member-authorization-based restoration, and cross-model memory aggregation, without requiring man-in-the-middle decryption of the terminal's official application. For sessions that connect directly to the model via the terminal's official application without going through the aggregation entry point, this system only manages the connection level and does not advocate for content-level anonymization or downlink restoration. Second, compliance basis: Cloud-based content governance is based on informed consent as a prerequisite for compliance. For minors, informed consent must be obtained by their guardians; for adult family members and visitors, informed consent must be obtained upon their first use of the cloud platform; for enterprise employees, governance in accordance with legally established labor regulations must be communicated in writing beforehand regarding the purpose, method, type, time limit, and exercise of rights, and is limited to what is necessary for human resource management; personal data traffic on employees' own devices is not within the scope of governance. Third, data minimization and immediate destruction: Cloud-based tenant governance points adhere to the principle of data minimization for content to be governed via the cloud path, deleting it immediately after processing without retaining the original text or using it for model training; storage is encrypted; deployment is localized according to the laws and regulations of the initial release region; and query, correction, deletion, and transfer methods are provided. For scenarios where the original value must not be visible in the cloud, the terminal is first anonymized before being uploaded to the cloud, or processed in a trusted execution environment where the customer controls the key; the two options are selected according to the session level. Fourth, privacy settings are not silently downgraded: Switching from local to cloud is a change in processing method. When switching, the system will explicitly prompt that the current privacy settings have changed due to cloud governance, and allow users to pause or request to return to local mode with one click. (4) Deterministic mapping of privacy protection levels like Figure 5 As shown, the two governance points apply the same set of membership policies for the same session type, but the location of the original data is determined by the governance point. When passing through the local governance anchor point, the original sensitive value identified by the uplink and the downlink restoration key are only stored in the sealed area corresponding to the member within the local trust root, and do not enter the forwarding plane or upload to the cloud. When passing through the cloud tenant governance point, the technical prerequisites and compliance boundaries in Section (3) shall be followed; Before switching governance points, it is first verified whether the target governance point can meet the privacy level required by the member policy for the session. Whether the target governance point can meet the required privacy level refers to whether it can provide content-level governance with the isolation and non-retention conditions required by the member policy, including whether it has the agreed trusted execution environment, terminal anonymization, or equivalent isolation boundaries. If the cloud tenant governance point does not meet the above conditions and cannot meet the required level, the session will not be allowed to a lower privacy level, but will be redirected to the local governance anchor point via the backhaul channel or processed only locally according to the preset. Only when backhaul and local processing only are not feasible will the flow be controlled and disconnected. The reason for the flow disconnection and the optional processing methods such as backhauling to local or local processing only will be explicitly prompted on the management terminal or access bearer program for the user to choose from, so as not to reduce the privacy protection level without the user's awareness. (5) Integration of idempotent measurement ledgers and memory like Figure 3As shown, the local governance anchor point and the cloud tenant governance point share the same metering ledger and the same member memory domain. Each AI target session is assigned a globally unique session identifier within the account by this system upon establishment. This identifier uses either a globally unique identifier or a monotonically increasing session sequence number, and does not use session identifiers assigned by the model service or the official terminal application. The idempotent key is generated by concatenating the member identifier, device identifier, and session identifier, and then extracting a digest, intentionally excluding timestamps. Each shard of a streaming session carries a monotonically increasing shard sequence number. During offline periods, both points accumulate metering records and memory entries respectively. After connection restoration, duplicate entries are deduplicated and merged using the idempotent key, retaining the initial record. Memory entries generated during cloud governance by members are temporarily stored in the cloud after end-to-end encryption using a key derived from the member's terminal key or local root of trust, unknown to the cloud. When the account is in pure cloud mode, this key is derived from the member's terminal key and decrypted by the member's terminal. The member's terminal key is generated and sealed in the terminal's secure storage by the access bearer program. In pure cloud mode, its root of trust is a protected key library provided by the terminal operating system or a terminal hardware security module. After the account is bound to a local governance anchor point, the ciphertext is repackaged in the member's terminal using the anchor point's sealed public key and migrated for protection using a key derived from the local root of trust. The plaintext does not leave the terminal. After the member's terminal or its anchor point retrieves the ciphertext, it is decrypted in the member's terminal or local root of trust and written into the member's memory domain according to the memory trust level. The cloud does not permanently store the plaintext memory. The trust level of memory entries, the writing threshold, user error correction, and verifiable deletion are carried out by the cross-large model memory layer technical solution that is compatible with this application. (6) Three-state governance markers, unknown status and discovery of unmanaged terminals The system outputs a governance status flag for each session: those governed by the local anchor point are locally governed, those governed by the cloud governance point are cloud governed, and those neither are governed are ungoverned. When a member terminal accesses artificial intelligence services via its own cellular baseband to access fourth-generation, fifth-generation, and subsequent generations of mobile communication networks, its session does not physically pass through the local anchor point. This ungoverned status is identified and reported by the access bearer program on the terminal and displayed on the management terminal. If the access bearer program is closed, bypassed, or unable to report due to operating system restrictions, the management terminal will truthfully present the terminal as having an unknown status for that period. The unknown status is based on the absence of evidence generated by the registered terminal within a set time period, where neither the target AI session passes through the anchor point nor the access bearer program reports online heartbeats or reachability. It is independent of the three governance flags of local governance, cloud governance, and ungoverned, indicating that the terminal is not visible to the system during that period, but does not indicate whether a target AI session occurred during that period. Cloud-based tenant governance points or local anchor points can also generate a notification for terminals to be protected based on terminals within the local network that have initiated AI target sessions but have not yet installed the access bearer program and have not been assigned to any member. This notification is based on the premise that the terminal has actually generated an AI target session, and does not scan or mark terminals that have not generated AI target sessions. After the administrator confirms on the management terminal, the binding entry point of the access bearer program is pushed to the terminal. (7) Lossless upgrade and downgrade between pure cloud form and bound anchor point form like Figure 4 As shown, the account exists in two stable forms. In the pure cloud form, the account is not yet bound to a local anchor point. Members terminate the AI ​​target session at the cloud governance point through the access bearer program. The member directory, quotas, and ledger are maintained by the cloud, and the memory is stored in the form of end-to-end encrypted ciphertext or desensitized digest. When bound to a local anchor point, the cloud sends the member directory, member policy version, quota balance, metering ledger, and end-to-end encrypted memory ciphertext to the anchor point. For the historical memory ciphertext originally protected by the member terminal key, when the member terminal or administrator terminal is present, the terminal decrypts it with its terminal key, and then re-encapsulates it with the sealed public key derived from the anchor point's local trust root before handing it over to the anchor point. During the re-encapsulation process, the plaintext does not leave the terminal trust boundary, and the cloud only sees the ciphertext. The anchor point derives mutually isolated sealed keys for each member using the local trust root and establishes a sealed policy instance. The member identifier, quota balance, session, and memory history are continuous before and after binding. Subsequently, when the member policy is changed in the cloud, the cloud pushes an incremental update of the policy version without the original sensitive value to the anchor point. When an account is unbound from the anchor point, or the anchor point fails or is offline for a long period of time, the cloud continues to govern in a pure cloud form according to the policy version and ledger of the last synchronization; the original value and restoration key in the local root of trust are not migrated to the cloud, only the end-to-end encrypted ciphertext or digest without the original value is allowed to be migrated, and the user can choose to keep or delete the cloud copy. (8) Administrator authorization options include cloud and governance point selection. The administrator authorization options mentioned in this section are available on the cloud, provided that the account is bound to a local governance anchor point and the member device is within the roaming governance system of that anchor point; pure cloud accounts that are not bound to a local governance anchor point are not within the scope of this section. like Figure 6 As shown, after an account is bound to a local governance anchor point, the default routing policy switches to local priority, and member terminals will default to local governance anchor point governance when they are outside the local network. Administrators can configure a switch in the management console to "Allow ordinary users to choose cloud governance when outside the local network." • When the switch is off (default state), all members will reconnect to the local governance anchor point when they are away from the cloud, and ordinary users do not have the right to choose the cloud themselves; • After the switch is turned on, ordinary users can choose "go to the cloud" or "go back to local" in the access bearer program; • Administrators can also configure differentiated permission policies for specific members, specific devices, or specific time periods; • High-restriction members are not allowed to choose the cloud themselves and are always forced to connect back to the local machine; After a regular user selects the cloud, their AI target session is routed to the cloud tenant governance point for governance. The cloud tenant governance point is the second governance point for the family or organization account, and content-level governance is implemented according to the technical premises and compliance boundaries in Section (3). The metering records and memory entries of sessions governed by the cloud and sessions governed locally are reported with idempotent keys composed of member identifier, device identifier and session identifier, respectively, and are merged into the same metering ledger and the same member memory domain; Regarding the fallback strategy for highly restricted members, this invention distinguishes between two categories: The first category is minors, whose fallback strategy is preset by their guardians at the management end, selected from allowing access via enhanced filtering cloud or controlled disconnection when local access is unavailable; the second category is personnel bound by confidentiality obligations, whose fallback strategy is not to fall back to the ordinary cloud, but to use terminal-level desensitization or client-controlled key trusted execution environment processing, and if neither is feasible, to handle it with strict snapshot processing or controlled disconnection. Among them, terminal-level desensitization means that the sensitive original value is replaced with pseudonyms or placeholders at the member's terminal before being sent up, and the cloud tenant governance point does not hold the restoration key or perform downlink original value restoration, but can still perform content-level security filtering and measurement on the desensitized content without relying on the original value; the client-controlled key trusted execution environment path performs complete content-level governance on the encrypted text within the trusted execution environment; (9) Access bearer program, credential system and compliance billing The access bearer or local anchor point identifies AI-targeted sessions based on AI service target data and directs them to the corresponding governance point. Non-AI-targeted sessions are not redirected, terminated, or have their paths altered. The primary access credential for a family or organization's model service is registered by the administrator in the cloud or at the anchor point. Both points share the same credential system and generate virtual access credentials for members, limiting the model scope and quotas. Members do not access the primary access credential. The model service, according to its open interface's metering rules, identifies this system as a client application, identifies members as end users, and charges based on token usage. Multiple family members or organization members under one account accessing through the same client application and being charged based on token usage are considered end-users allowed by the open interface. Member virtual access credentials are restricted access tokens or credential references that limit the scope, validity period, and quota of the model. During synchronization between the cloud and local anchors, they are encapsulated with the sealed public key of the receiving point. When the receiving point is a local governance anchor, its local root-of-trust derived public key is used; when the receiving point is a cloud tenant governance point, the tenant-encapsulated public key of that account under the cloud key system is used. They are not transmitted in plaintext as raw credentials that could bypass the local governance anchor and directly access the model service. End-applications establish sessions using their own credentials, and the anchor cannot replace these credentials at the network layer. These sessions are only managed at the connection level, and member virtual access credentials are not injected into them. Cloud tenant governance points initiating requests to the model service on behalf of members only occur in pure cloud scenarios or sessions terminated through an aggregation entry point, and the credentials are used within a sealed execution environment or equivalent isolation boundary. (10) Differences from existing plans and supporting plans Compared to cloud-based multi-model aggregation services, the cloud-based tenant governance point of this invention is not the sole execution point. Within the broadband entry point, there exists a local governance anchor point with a local root of trust as its authority. This can cover all terminals within the home that have not installed cloud programs or configured cloud credentials, and provides a lossless upgrade path from pure cloud to physical boxes. Compared to enterprise-level AI gateways that support multiple deployment models such as Software as a Service, Virtual Private Cloud, and self-hosting, this invention targets homes and small organizations without dedicated maintenance personnel. Its local end is a plug-and-play broadband entry point anchor point holding a restoration key via a hardware root of trust, rather than a customer-managed cluster. It provides consumers with lossless inheritance from zero-hardware pure cloud applications to QR code-bound anchor points within the same account. Compared to home network security boxes that work with mobile applications, the cloud in this invention is a second governance point, equally responsible for AI conversation governance as the local anchor point. Between the two points, there are form- and presence-related routing, privacy tier mapping, idempotent ledger memory fusion, and three-state marking. Compared to enterprise security access service edge and zero-trust solutions, the governance authority of this invention is based on the local trust root of the family or organization, without relying on resident clients, enterprise identity systems and dedicated operation and maintenance, and supports a pure cloud starting point without anchor points. It is necessary to further clarify the boundary between the present invention and the supporting technical solutions: the specific processing of two-way streaming reversible desensitization of uplink sensitive information, placeholder deformation, and downlink restoration according to member authorization level is carried by the session desensitization and restoration technical solution supporting this application; the artificial intelligence service identification and deterministic binary classification of encrypted traffic without decryption is carried by the encrypted traffic governance technical solution supporting this application; the trusted grading of writing across large model memories, anti-pollution write gates, injection isolation and sub-account memory domains are carried by the cross-large model memory layer technical solution supporting this application; the specific path of the member's pre-signed roaming credential when out of home, zero-knowledge blind relay back to the local anchor point and home idempotent takeover is carried by the session roaming governance technical solution supporting this application; the forwarding and governance isolation of the entry device, and the failure bypass path are carried by the in-transit governance integrated access device technical solution supporting this application. The present invention does not repetitively claim the above specific processing, and its inventive point lies in the cloud-edge seam of the two peer governance points (the cloud tenant governance point and the local governance anchor point under the same family or organization account) in terms of presence-related routing selection, privacy gear deterministic mapping, idempotent fusion of metering ledger and memory, governance status marking, lossless upgrade and downgrade of accounts between the pure cloud form and the anchor-bound form, as well as the administrator-configurable local priority and cloud optional governance point selection mechanism after the account is bound to the anchor point; (11) Exemplary Embodiments Embodiment 1, box-free pure cloud start-up: A family has not purchased a local anchor point yet. The parent installs the access bearing program on the mobile phone and tablet, registers a family account, adds two members (a child and an adult) and sets the content level and quota for each respectively. The artificial intelligence target sessions of the two members are governed by the cloud governance point according to their respective policies, the memories are temporarily stored with end-to-end encryption using the parent terminal key, and the metering is recorded in the same family ledger; Embodiment 2, lossless upgrade after purchasing the anchor point: The family subsequently purchases a local governance anchor point and connects it in series between the optical modem and the router. The parent scans the code on the management end to bind the anchor point to the original family account, and the cloud issues the member directory, policies, balance and encrypted memories. The historical memory ciphertext is decrypted at the parent terminal with the parent terminal key, re-encapsulated with the anchor point's sealed public key and then delivered to the anchor point, the plaintext does not leave the terminal, and the anchor point establishes sealed instances for the two members with the local trust root; smart TVs, smart speakers and visitor terminals are managed by the local anchor point without installing programs, and the child's original quota balance, dialogue and memory history are continuous; after binding, the default routing policy is switched to local priority; Embodiment 3, seamless switching between home and away: When the child uses the tablet at home, the session is locally governed by the anchor point according to the child's policy, and the original value does not leave the home; after going out with the parent, the tablet connects to a public wireless hotspot, the presence is determined to be outside the coverage of the local governance anchor point, since the child is a preset high-restriction member, its artificial intelligence target session is connected back to the home anchor point through the backhaul channel to continue to be governed according to the same child policy, and ordinary Internet traffic does not connect back; after returning home, the session is idempotently taken over by the anchor point without repeated billing; Example 4: Cloud Access Option After Administrator Authorization: An employee of a company is on a business trip. The administrator has enabled the "Allow ordinary users to choose cloud access when away from home" switch on the management terminal. After the employee selects "Go to cloud access" while away from home, their AI target session terminates at the aggregation entry point of this system, is forwarded to the model service after content-level governance, and is subject to compliance with the requirement of informed consent in the cloud. The cloud is used and destroyed without retention or training. Measurement and memory are merged into the same ledger and memory domain with idempotent keys. After returning home, the encrypted memory is retrieved from the anchor point and merged into the local memory domain. Example 5: Forced Local Connection for High-Restriction Members: When a minor member is away, the system automatically identifies them as a high-restriction member, does not allow them to choose the cloud independently, and always forces them to connect back to the local governance anchor point, and governs them according to child-level strategies; if the local connection is unreachable, the system selects to allow access to the enhanced-filter cloud or control the disconnection according to the strategy preset by the guardian on the management terminal. Example 6: Confidential positions do not fall back to the ordinary cloud: When a partner of a law firm is on a business trip, the system identifies him as a person bound by confidentiality obligations. When the local anchor point is unreachable, the system does not fall back to the ordinary cloud, but uses terminal desensitization or a trusted execution environment where the client controls the key. If neither of these is feasible, the system takes a strict snapshot or cuts off the flow under control. Example 7: Cellular disconnection and unknown status: When a child goes out, turns off Wi-Fi and uses cellular data, and asks a question directly in the official large model application. If the access bearer is still running, the session is marked as an unmanaged out-of-home session and a notification is sent to the parent's mobile management terminal. If the access bearer is shut down by the system and cannot be reported, the management terminal only gives an unknown status gap notification for the offline period of the terminal, without specifying whether it accesses artificial intelligence services. Example 8, Anchor Point Failure and Unbinding: During the anchor point repair period, the cloud continues to govern in a pure cloud form according to the last synchronized strategy and ledger. The original value and restoration key remain in the anchor point's local trust root and are not uploaded to the cloud. After the family decides to permanently disable the anchor point and unbind it, the account retains the member directory, ledger and encrypted memory to continue working in a pure cloud form. Example 9: Small Organizations with Multiple Locations and Version Alignment: A small design studio deploys an anchor point at its office and at the founder's residence, both bound to the same organizational account. Employees at the company are managed according to their job positions and data hierarchy, while those at the founder's residence connect based on accessibility. Both anchor points share the same organizational ledger with the cloud, synchronizing only policy versions, member pseudonyms, quota counts, and status markers between anchor points. If the policy version of the residence anchor point is outdated and the employee's job position falls within the version difference range, the session will either incrementally update the policy or switch back to the main anchor point at the office. Example 10: Local Cloud Switching Without Silent Degradation: When an adult member uses a cloud path while outside, the system will display a prompt on the access host program interface indicating that the current cloud governance and privacy settings have changed, and provide one-click pause or request to return to local; after local access is restored, it will automatically switch back to local, with no silent degradation throughout the entire process.

Claims

1. A unified account-based AI session governance method, characterized in that, The process is executed by an account collaborative governance system, which includes a cloud-based tenant governance point, an access bearer program on member terminals, and a local governance anchor point selectively bound to accounts within the same family or organization. The local governance anchor point is deployed at the broadband entry point of the family or organization, sealing the member-isolated policy instance and restoration key with a local root of trust. The cloud-based tenant governance point maintains a unique member directory, quota, and metering ledger for the account. The member-based policy includes two levels: connection level and content level. The cloud-based tenant governance point and the local governance anchor point apply the corresponding level based on whether their session can obtain application-layer plaintext. The method includes: the account is not bound... When a local governance anchor point is defined, the system is in a pure cloud environment. The AI ​​target session of a member terminal terminates and accepts governance at the cloud tenant governance point via the access bearer program. After an account is bound to a local governance anchor point, the cloud tenant governance point synchronizes the member policy version and member pseudonym identifier (excluding original sensitive values) from the member directory to the local governance anchor point. The local governance anchor point then becomes the local governance point for that account within its broadband entry point. When a member terminal initiates an AI target session, the access bearer program or the local governance anchor point determines its presence based on the reachability of the local governance anchor point. If the terminal is within the local network, it is routed to the local governance anchor point for governance. If the terminal is outside the local network, and the account... If the account is in pure cloud mode, it will be routed to the cloud tenant governance point by default. If the account is bound to a local governance anchor point, it will be routed back to the local governance anchor point by default, or cloud governance will be selected according to the policy pre-configured by the administrator. When governed by the local governance anchor point, the original sensitive values ​​and restoration keys do not leave the local root of trust. When governed by the cloud tenant governance point, the content to be governed terminates at the aggregation entry point of this system, and the system connects to the terminal and model service respectively through two independent connections. The application layer content is naturally visible at the aggregation entry point. The metering records and memory entries generated by members at the two governance points are reported with idempotent keys composed of member identifier, device identifier and session identifier, and merged into the same metering point. For the same member memory domain, repeated reporting of the same idempotent key will not result in repeated metering or solidification. For each session, a governance status flag will be output: those governed by the local governance anchor will be marked as local governance, those governed by the cloud tenant governance point will be marked as cloud governance, and those not governed by either will be marked as ungoverned. This status will be displayed on the management end by member. After an account is unbound or removed from the local governance anchor, the member directory, quota, metering ledger, and encrypted memory will be retained at the cloud tenant governance point. The account will revert to pure cloud mode and continue working. When a pure cloud mode account is bound to a local governance anchor for the first time, its member directory, member-based policy, metering ledger, and encrypted memory will be inherited to the local governance point without loss.

2. The method according to claim 1, characterized in that, The presence determination includes at least two types of evidence: evidence that the access bearer program receives periodic reachability identifiers declared by the local governance anchor within the local network; evidence that the terminal and the local governance anchor are in the same address prefix or the same link layer broadcast domain; and online status reported by the local governance anchor to the cloud tenant governance point via an outbound long connection. When the at least two types of evidence consistently indicate that the terminal is within the coverage area of ​​the local governance anchor, it is determined to be present. If the evidence is insufficient or contradictory, it is determined to be outside the coverage area of ​​the local governance anchor. For terminals without the access bearer program installed, the local governance anchor does not rely on the access bearer program to receive periodic declarations, but independently completes the presence determination based on two types of network layer evidence: the terminal and the anchor are in the same link layer broadcast domain or the same network address prefix, and the terminal's AI target session physically transits through the anchor.

3. The method according to claim 2, characterized in that, When a terminal is determined to be outside the coverage area of ​​the local governance anchor point, the access bearer program determines the route according to the account type, member policy and user selection: for pure cloud accounts, the AI ​​target session is routed to the cloud tenant governance point; for accounts that have been bound to the local governance anchor point, the AI ​​target session is by default routed back to the local governance anchor point through the back-call channel established with the session roaming governance solution that is compatible with this application. For users with pre-defined high-restriction members, sessions that require access to the member's local memory domain, or sessions where users explicitly select strong consistency governance, the local governance anchor point will be forcibly reconnected. Non-AI target sessions will not reconnect. If the local governance anchor point is unreachable, the session will be processed according to the strict policy snapshot pre-signed before the anchor point went offline, or the AI ​​target session will be routed to the cloud tenant governance point for governance when the administrator pre-authorizes it. High-restriction members will be processed according to the fallback method pre-defined by their member policy. If none of these conditions are met, the session will be marked as unmanaged.

4. The method according to claim 1, characterized in that, When a terminal directly accesses a fourth-generation, fifth-generation, or later-generation mobile communication network via its own cellular baseband, and its AI target session does not physically pass through a local governance anchor point, the access bearer program on the terminal marks the session as an ungoverned outbound session and prompts the management terminal. When the access bearer program is closed, bypassed, or unable to mark the session due to operating system limitations, the management terminal truthfully presents the terminal as having an unknown state for that period. This unknown state is independent of the three types of session governance marking: local governance, cloud governance, and ungoverned. It is based on the absence of evidence generated by the registered terminal within a set time period, indicating that the terminal is not visible to the system during that period, but does not imply whether an AI target session occurred during that period. When a terminal is on the local network but bypasses the local governance anchor point through an unregistered encrypted tunnel, the local governance anchor point marks the session as ungoverned. After the terminal is authorized to reconnect to the local governance anchor point or access the cloud tenant governance point, the sessions from that point onward are marked as either locally governed or cloud governed, respectively. This change in marking only changes the classification of the governance status and does not retroactively govern sessions that have been completed before the marking.

5. The method according to claim 1, characterized in that, The session identifier is assigned by the account collaborative governance system when the AI ​​target session is established. It is globally unique within the account and uses a globally unique identifier or a monotonically increasing session sequence number. It does not use session identifiers assigned by the model service or terminal application itself. The idempotent key is generated by taking a digest after fixedly concatenating the member identifier, device identifier and the session identifier. The idempotent key does not contain a timestamp. Each fragment of the same session carries a monotonically increasing fragment sequence number. The local governance anchor point and the cloud tenant governance point accumulate metering records and memory entries respectively during offline periods. After the connection is restored, the idempotent key is used to deduplicate and merge the records while retaining the first record. This ensures that the same AI target session is not repeatedly billed or repeatedly stored due to switching between two governance points.

6. The method according to claim 5, characterized in that, Memory entries generated by members during the management of cloud tenant governance points are temporarily stored in the cloud after end-to-end encryption using a key derived from the member's terminal key or local trust root and unknown to the cloud tenant governance point. When the account is in pure cloud mode, the key is derived from the member terminal key. After the account is bound to a local governance anchor, the key can be derived from the local root of trust. After the member terminal or its local governance anchor retrieves the ciphertext, it is decrypted in the member terminal or local root of trust and written into the member's memory field according to the memory trust level. The cloud tenant governance point does not fix the member's plaintext memory.

7. The method according to claim 1, characterized in that, When a pure cloud account is first bound to a local governance anchor, the cloud tenant governance point sends a member directory, member policy version, quota balance, metering ledger, and end-to-end encrypted memory ciphertext to the anchor. The local governance anchor derives isolated sealing keys for each member using the local root of trust and establishes a sealing policy instance. Member identifiers, quota balances, sessions, and memory history are continuous before and after binding, and members do not need to re-register or reconfigure agents on member terminals. If a member policy changes after binding, the cloud tenant governance point pushes an incremental update of the policy version without the original sensitive values ​​to the bound local governance anchor, and the local governance anchor updates its sealing policy instance accordingly.

8. The method according to claim 7, characterized in that, When an account is unbound from the local governance anchor, or when the local governance anchor fails or is offline for an extended period exceeding the set time, the cloud tenant governance point continues to govern in a pure cloud manner based on the last synchronized member policy version and metering ledger. The original sensitive values ​​and restoration keys in the local root of trust are not migrated to the cloud tenant governance point. Only end-to-end encrypted ciphertext or digests without the original values ​​are allowed to be migrated to the cloud, and the user can choose whether to retain or delete the cloud copy.

9. The method according to claim 1, characterized in that, The access bearer program can be at least one of web applications, native applications, operating system proxies, or lightweight clients. The access bearer program or local governance anchor point only identifies AI target sessions based on AI service target data and directs them to the corresponding governance point. Non-AI target sessions are not redirected, terminated, or have their network paths changed. For terminals that have not installed the access bearer program, when the account is bound to the local governance anchor point and the terminal is within the coverage area of ​​the local governance anchor point, the local governance anchor point determines the presence at the network layer and performs connection-level management on the session based on the fact that the AI ​​target session of the terminal is transiting within its local link layer broadcast domain or network address prefix. The connection-level management includes identifying AI target sessions, access control, concurrency and frequency management, usage metering, and connection-level handling according to visitor policies or administrator preset default policies. This handling is completed at the moment the anchor point forwards the session, without requiring the terminal to install programs, configure proxies, or trust certificates. For connection-level managed sessions, no application-layer global session identifier is assigned, and they are not included in content-level idempotent key deduplication. Instead, they are aggregated and measured according to at least one connection-level metric, such as target service, time window, number of connections, or byte volume, and the measurement of visitors or administrator-preset default members is included. For sessions on such terminals that are end-to-end encrypted at the application layer or have fixed certificates, and whose anchor points cannot obtain application-layer plaintext, content-level desensitization and downlink restoration are not recommended. The aforementioned content-level desensitization, downlink restoration according to member authorization level, and unified credential forwarding are only implemented when the AI ​​target session terminates at the aggregation entry point provided by this system and is connected to the terminal and model service by this system through two independent connections.

10. The method according to claim 1, characterized in that, The primary access credential for a family or organization's model service is registered by the administrator at either the cloud tenant governance point or the local governance anchor point. The cloud tenant governance point and the local governance anchor point share the same credential system and generate member virtual access credentials that limit the model scope and quota based on the member. Members do not access the primary access credential. The model service identifies the account as a client application and the member as an end user according to the metering rules of its open interface, and charges based on token usage. Member virtual access credentials are synchronized between the two governance points, with shared quotas and unified ledgers. The member virtual access credential is a restricted access token or credential reference that limits the model scope, validity period, and quota, and it exists at both governance points. During synchronization, the public key of the receiving point is used for encapsulation. When the receiving point is a local governance anchor, its local root of trust derived public key is used. When the receiving point is a cloud tenant governance point, the tenant encapsulated public key of the account under the cloud key system is used. Plaintext transmission of raw credentials that can bypass the local governance anchor and directly access the model service is not used. The terminal application establishes a session with its own credentials, and the anchor cannot replace the credentials at the network layer. It is only managed at the connection level according to claim 9, and no virtual access credentials for members are injected into it. The cloud tenant governance point initiates requests to the model service on behalf of members only in pure cloud mode or sessions terminated through the aggregation entry, and uses credentials within the sealed execution environment or equivalent isolation boundary.

11. The method according to claim 1, characterized in that, When switching governance points, the system first verifies whether the target governance point can meet the privacy protection level required by the member policy for the session. If the cloud tenant governance point cannot provide content-level governance with the isolation and non-retention conditions required by the member policy, such as not having the agreed trusted execution environment, terminal anonymization, or equivalent isolation boundary, the session will not be allowed to a lower privacy level. Instead, it will be redirected to the local governance anchor point via the callback channel established with the session roaming governance solution that is compatible with this application, or processed only locally. Only when both callback and local processing are not feasible will the flow be cut off in a controlled manner. The system will also explicitly prompt the management terminal or access bearer program about the changes in the current governance point and privacy level, and provide an operation entry for one-click pause or request to return to local, so as not to reduce the privacy protection level without the user's awareness.

12. The method according to claim 4, characterized in that, Cloud-based tenant governance points or local governance anchors generate protection prompts for terminals that have initiated AI target sessions but have not yet installed the access bearer program and have not been assigned to any member within the local network. Terminals that have not initiated AI target sessions are not scanned or marked. Guest terminals can choose not to bind and continue to be managed according to the guest policy. After the administrator confirms on the management terminal, the binding entry of the access bearer program is pushed to the terminal. After the binding is completed and the terminal is assigned to a member, the AI ​​target sessions of the terminal are included in the three-state marking and metering ledger of local governance, cloud governance, and ungoverned.

13. The method according to claim 1, characterized in that, When a family or organization account is bound to multiple local governance anchors deployed in different locations, member terminals reconnect to their primary anchor or the nearest anchor within the same organization based on reachability. Each local governance anchor and the cloud tenant governance point share the same member directory and metering ledger. Only policy version numbers, member pseudonym identifiers, quota counts, and session status markers are synchronized between anchors and between anchors and the cloud. Original sensitive values ​​and restoration keys do not flow across anchors or upload to the cloud. Before accepting the reconnection of the nearest anchor, the policy version number of its sealing policy instance is checked to see if it is consistent with the latest version. If the members or categories involved in the session are not within the version difference range, the local sealing policy instance is used for governance and incremental synchronization is performed in the background. If they are within the version difference range, incremental updates are completed first, or the reconnection to the primary anchor or cloud tenant governance point is changed for governance. Metering dimensions, metering units, and quota deduction rules are ledger rules that do not contain original sensitive values ​​and are always based on the unified metering ledger caliber maintained by the cloud tenant governance point.

14. The method according to claim 1, characterized in that, After the account is bound to the local governance anchor point, the default routing policy is switched to local priority. When the member terminal is outside, the default routing is back to the local governance anchor point for governance. After being pre-authorized by the administrator, ordinary users can choose to route the AI ​​target session to the cloud tenant governance point for governance. Ordinary users who choose the cloud need the administrator to enable the allow switch on the management terminal or authorize it one by one. High-restriction members are not allowed to choose the cloud on their own and are always forced to back to the local governance anchor point.

15. The method according to claim 14, characterized in that, The administrator can configure differentiated permission policies for specified members, devices, or time periods on the management terminal. When the permission switch is off, all members will reconnect to the local governance anchor point when they are outside, and ordinary users do not have the right to choose the cloud. When the permission switch is on, ordinary users can choose to go to the cloud or return to the local network when accessing the bearer program. The administrator's authorization to choose the cloud does not affect the governance authority and adjudication source status of the local governance anchor point. The cloud tenant governance point is only used as the second governance point for the account.

16. The method according to claim 14, characterized in that, The highly restricted members are divided into two categories: the first category is minors, whose fallback strategy is preset by the guardian on the management end, and is selected from allowing access to the enhanced-filter cloud or controlled disconnection when the local area is unreachable; the second category is personnel bound by confidentiality obligations, whose fallback strategy is not to fall back to the ordinary cloud, but to adopt terminal de-identification or trusted execution environment processing with client-controlled keys. When the terminal is de-identified, the cloud tenant governance point does not hold the restoration key and does not restore the downlink original value, but only performs content-level security filtering and measurement on the de-identified content without relying on the original value. If neither of the above is feasible, the stricter strategy is applied to snapshot processing or controlled disconnection.

17. An AI-powered session management device with unified account functionality, characterized in that, The device includes a cloud-based tenant governance point, an access bearer program on a member terminal, and a local governance anchor point selectively bound to an account of the same family or organization. When the local governance anchor point is bound, it seals the policy instance and restoration key of isolation by member with its local root of trust. The device is equipped with an account and member directory unit, an presence determination and routing unit, a two-point policy and privacy grading unit, an idempotent ledger and memory fusion unit, a governance status marking unit, an anchor point binding and upgrade / downgrade unit, and an administrator authorization and governance point selection unit. It is configured to perform the method of any one of claims 1 to 16.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method of any one of claims 1 to 16, wherein the computer-readable storage medium includes at least one of a cloud server storage medium, an access device firmware storage medium deployed at a broadband gateway, or a member terminal storage medium, and the access device includes at least one of a router, an optical network terminal converged gateway, a customer front-end device, a wireless access point, a network switch, network auxiliary storage, a soft router host, or a containerized instance.